Table 2.

Comparison of model to test experts neuroradiologists B and C on test sets.

Categories	Metric ${}^1$	Expert B					Expert C					Expert A
		Inter-expert ${}^2$ (B to A)		Model-expert${}^2$ (B to model)		p-value${}^3$ for non-inferiority	Inter-expert${}^2$ (C to A)		Model-expert${}^2$ (C to model)		p-value${}^3$ for non-inferiority	Model-expert${}^2$ (A to model)
Volume	VS	0.66	± 0.1	0.81	± 0.1	p<0.001	0.64	± 0.3	0.51	± 0.3	p<0.01	0.67	± 0.14
	AVD [ml]	8.40	± 5.25	7.11	± 4.81	Non-sig	7.28	± 4.96	5.99	± 2.24	p<0.05	7.43	± 4.31
Overlap	Dice	0.47	± 0.16	0.56	± 0.18	p<0.0001	0.25	± 0.15	0.36	± 0.15	p<0.0001	0.47	± 0.13
	Precision	0.49	± 0.26	0.52	± 0.18	p<0.0001	0.64	± 0.16	0.77	± 0.15	p<0.001	0.58	± 0.26
	Recall	0.59	± 0.18	0.73	± 0.16	p<0.0001	0.17	± 0.15	0.26	± 0.14	p<0.0001	0.52	± 0.15
Distance	HD 95 [mm]	15.89	± 5.02	12.39	± 3.78	Non-sig	21.97	± 7.36	18.13	± 7.03	Non-sig	18.04	± 9.21
	SDT 5 mm	0.54	± 0.09	0.63	± 0.16	p<0.0001	0.31	± 0.14	0.31	± 0.18	p<0.0001	0.46	± 0.09

${}^1$ VS volumetric similarity, AVD absolute volume difference, HD 95 Hausdorff distance 95th percentile, SDT surface dice at tolerance 5 mm ${}^2$ Median ± 95% CI (bootstrapped) ${}^3$ p-values of one-sided Wilcoxon sign rank test.